Co-Folding Confidence Dispersion Is a Coarse Binder Filter, Not an Affinity Ranker: Boltz-2 sigma-iptm Evaluated Against Potency Rank, Same-Chemotype Weak Binders, and Off-Chemotype Decoys on Matrix Metalloproteinase-1

Correction (new version, 2026-08-12). Version 0.4 corrects version 0.3. The off-chemotype sigma_E filtering result is withdrawn on both potentials: under nearest-neighbour property matching its AUC falls from 0.693 to 0.574 [0.399, 0.734] (GFN2-xTB) and from 0.682 to 0.584 [0.425, 0.738] (Orb-v3), on intervals that include chance. The sigma_iptm discrimination result is unaffected and is strengthened under the same controls: 0.993 [0.969, 1.000] on closest matched pairs, with the active lower in 52 of 52 pairs. The term "property-matched" is replaced throughout, including in the title, by "off-chemotype (ECFP4 Tanimoto < 0.35)", which is what the cohort construction actually guarantees; the arms are separable on bulk physicochemistry alone at AUC 0.917, so every discrimination figure on that cohort is now reported against that baseline rather than against 0.5. Two new subsections report the cohort-construction controls (3.11.1) and an out-of-fold descriptor baseline on the hard cohort (3.11.2), where five descriptors computed from the SMILES string reach 0.822, 0.896 and 0.860 against sigma_iptm's 0.698, 0.751 and 0.756 and sigma_iptm adds nothing on top of them. The licensing statement is narrowed accordingly: sigma_iptm is licensed against off-chemotype negatives, not at the same-chemotype end, and the 0.68 figure is retained as a measured bound on the gradient rather than offered as an operating point. A further subsection (3.3.2.1) reports twelve runs confirming that the CHEMBL259829 sigma_iptm outlier is not inter-run determinism, and records that the magnitude of sigma_iptm is not portable across cofold configurations although its ordering is. The dispersion of co-folding confidence across an inference ensemble is widely used to triage protein-ligand predictions, but what it measures has not been fixed against experiment. This work evaluates Boltz-2 sigma-iptm, the standard deviation of iptm over 100 diffusion samples, on the Zn(II) metalloprotease MMP-1 across three cohorts of increasing negative-distance, all co-folded under one protocol. Among 93 ChEMBL actives spanning 3.62 log units of IC50 it does not rank potency: the Spearman correlation with the model's own potency-rank error is -0.005 (95% CI [-0.21, +0.20]), which bounds a small effect rather than excluding one. Among 121 PubChem compounds spanning 5.10 log units, 36 of them weaker than 10 micromolar, it separates the arms at every threshold tested (AUC 0.68 to 0.74, all CIs excluding 0.5; Spearman with pIC50 -0.443, essentially unchanged at -0.440 after controlling for heavy-atom count). Against 90 off-chemotype ZINC decoys (ECFP4 Tanimoto below 0.35) drawn against 52 sub-micromolar actives by nearest-neighbour matching on six descriptors, separation is near-perfect (AUC 0.992 [0.980, 0.999]; 0.993 [0.969, 1.000] on closest property-matched pairs). The resulting gradient, null then 0.68 to 0.74 then 0.99, is monotonic in how different the negatives are, which is the signature of a coarse confidence filter rather than of an affinity model. The energetic analog computed on identical poses, pose-ensemble sigma_E under GFN2-xTB and Orb-v3 OMol25, is weaker at every level despite the two potentials agreeing at r = +0.982. The measurement basis is a systematic 161-reseed audit of a 15-ligand panel that resolves per-ligand sigma and surfaces two orthogonal outlier classes in double dissociation: one ligand carries the panel's largest ligand-energy dispersion while ranking fifth of fifteen on confidence dispersion, and another is the largest confidence outlier while its energy dispersion is mid-panel. The framework is offered as a detection protocol; the upstream interventions it proposes for either axis are not validated in this work. sigma-iptm is therefore licensed against off-chemotype negatives, where it holds 0.993 under property matching, and not at the same-chemotype end, where five descriptors computed from the SMILES string reach 0.82 to 0.90 out of fold against its 0.70 to 0.76; the 0.68 figure is a measured bound on the gradient rather than an operating point, and sigma-iptm is not a potency ranker among binders. All cohort data, analysis code and persisted outputs are attached; the code repository is https://github.com/crazat/genesis_medicine (Apache-2.0). Provenance note: the 15-ligand reseed panel is used for structural reproducibility only. No compound identity or potency claim is made for any of its members, following the 2026-07-16 primary-source audit that withdrew a fabricated calibration panel from this project. Every potency figure in this record comes from the PubChem-sourced and ChEMBL-sourced cohorts, each row of which carries its own source identifier and retrieval timestamp in the attached files.